Short-Term Photovoltaic Power Forcasting Based On Deep Learning Algorithm

With the depletion of fossil energy and global climate change,the development of renewable energy such as solar energy has increased,and the proportion of photovoltaic power connected to the grid has also increased.The power of photovoltaic power generation is determined by the intensity of solar radiation.At the same time,it is affected by meteorological factors such as ambient temperature and humidity.It not only has strong randomness and volatility,but also has periodic and day and night alternation,which makes large-scale photovoltaic grid connection bring a huge impact on the stability of the power system.In order to reduce the impact of photovoltaic grid-connected power on the power system as much as possible,it is necessary to carry out short-term PV power forecasting.The forecast results can be used as a reference for grid dispatching and decision-making of PV power plant operation and maintenance.In order to further improve the accuracy of photovoltaic power generation prediction and reduce the artificial participation in data feature extraction,this paper analyzes the research status of photovoltaic power generation prediction technology and deep learning algorithm at home and abroad,and then relies on the University of Queensland photovoltaic power generation system for six consecutive years of operation data,analysis and research on the law of photovoltaic power generation,the main contents of the work are as follows:(1)The basic situation of photovoltaic power prediction technology is reviewed,and the existing problems of existing prediction methods are analyzed.Then take the photovoltaic power generation system of the University of Queensland in Australia as an example,analyze the influence of various meteorological factors on the photovoltaic power generation by plotting the variables,and then use the Pearson correlation coefficient formula to further determine the correlation degree.At the same time,analyze the feasibility of application using the deep learning algorithm for photovoltaic power forecasting,lay the foundation for subsequent research.(2)Introduce the basic concepts of artificial neural network and the basic principles of deep learning algorithms such as deep belief network,stacked auto-encoder,convolutional neural network,long short-term memory network,analyze each of their advantages,disadvantages and applicable field,and then The photovoltaic power prediction problem gives an improved solution.(3)Pre-processing and visual analysis of the original data set to ensure the reliability and integrity of the data for later use.(4)A short-term photovoltaic power generation prediction method based on LSTM deep learning model is proposed.First introduce the construction process of the model,and then determine the network structure and key parameters through experiments.Finally,using the example dataset to test the model,the prediction effect is summarized and compared,and proposed short-term photovoltaic power prediction method based on LSTM is better.(5)In order to comprehensively utilize the excellent performance of different deep learning models to further improve the accuracy and generalization ability of the PV prediction model,a short-term photovoltaic power generation prediction method based on combined deep learning model is proposed,taking the combined model CNN-LSTM as an example.Firstly,the structural parameters,input data forms and other key parameters of the model are determined experimentally.Then,the CNN-LSTM model is used to predict experimentally using the instance data,and the other two combined deep learning models and LSTM models are tested using the same data set.Compare the performance indicators of the model.The results show that the performance of the CNN-LSTM model is the best,which indicates that the short-term photovoltaic power generation prediction method based on the combined deep learning model proposed in this paper can effectively integrate the data feature extraction performance of CNN and the time series prediction performance of LSTM,thus obtains a better effect than a single model.